F and P values for random factors

Hi Mahendra,

You can assess the significance of both fixed and random factors using
likelihood ratio tests.  Say you want to test the significance of predictor
A in a model with predictors A and B and random slopes for A.  You can do:

full_model = lmer(y ~ A + B)
model_without_A = lmer(y ~ B)
anova(full_model, model_without_A)

The anova() function in this case will return a chi-squared score (df =
number of predic) and a p-value.  The same procedure can be used for random
effects, e.g.:

full_model_2 = lmer(y ~ A + B + (1+A | random_thing)
model_without_random_slope_for_A = lmer(y ~ A + B + (1 | random_thing)
anova(full_model, model_without_random_slope_for_A)

This works because the log-likelihoods of nested models, in the limit,
approximate a chi-squared distribution.

See:  https://en.wikipedia.org/wiki/Likelihood-ratio_test

Or maybe that wasn't what you were asking at all.

Also I think you forgot to attach the file.

Hope that helps!
Alex


On Thu, Feb 4, 2016 at 8:32 PM, Mahendra Dia <diamahendra at gmail.com>
wrote:

Hi.

I am reaching you out to learn how to compute F ratio and P values for my
experiment where all the factors are treated as random factors.
Please see the attached file where I explained my treatments and sample
data.

I thank you in advance.

Sincerely,
Mahendra-

--
Alex Fine
Ph. (336) 302-3251
web:  abfine.github.io/

        [[alternative HTML version deleted]]